Met Recommendation System
Gabriel Gianordoli and John Choi
MET Media Lab
Spring, 2015
Recommendation system based on the examples from Toby Segara's book Programming Collective Intelligence.
The data source is a list of items saved by users on their MyMet accounts. The files inside recommendation_item_based turn the records into a list of similar items, following this sequence:
- 01_make_dict_user_items groups the records by user: [ { user1: [ item1, item2, ... ], ... } ]
- 02_make_dict_item_users switches the collection above, grouping items based on the users who saved them: [ { item1: [ user1, user2, ... ], ... } ]
- 03_make_dict_similar_items runs the similarity algorithms and build a list of 10 similar items for each artwork in the collection: [ { item1: [ similar1, similar2, ..., similar10 ], ... } ]
- 04_json_create gets more information about each item using scrapi. The final format for each item is:
{
item_id: 16584,
item_title: 'George Washington',
gallery_number: 140/null, // null if object is not on display
department: 'egyptian art'/null,
img_url_web: 'http://...', // null if it doesn't have image?
img_url_big: 'http://...', // null if it doesn't have image?
similar_items: [
{ item_id: 004327, similarity: 0.01 },
{ item_id: 052345, similarity: 0.005 },
...
]
}
See more at the project description on the MET Media Lab's Hackpad